Comment 114 for bug 127101

The release notes are at https://wiki.ubuntu.com/GutsyGibbon/RC. I think it's a good suggestion to mention this issue in the Caveats section, in a short and constructive manner. Would you mind taking a shot at creating an initial draft of the text to go in here?

Given the plan of putting out a release note and fixing this in a post-release update, we need to approach this bug a little differently. First, we need to focus on finding "regression-free" fixes or workarounds. We will have made mention of the issue in the release notes, so it would not be good to put out a fix that risks just moving the bug to a new, unknown location. Second, I think we need to ensure when we apply a fix, we have a deep understanding of *why* it works; otherwise, we won't have confidence that we're truly fixing the root cause.

I posted about this issue to the xorg@ mailing list last night, to get advice. Unfortunately, there's not been very little response.

So far, the best attack angle we have are Juan Pablo's backtraces. These give tangible, deep information into the issue, that could possibly identify a real solution. I'm going to be focusing on this angle. I would encourage others who wish to help solve this issue to follow Juan Pablo's approach and try to also gain backtraces like his.

unggnu, thanks for investigating the angle of removing the overlay fix. However it sounds like that is more a fix for bug 141063 than this one. Also, since presumably removing the fix would result in returning the original bug that it fixed, it couldn't be considered regression-free.

Jose, it's quite interesting that enabling EXA seems to reduce the frequency of the issue, and that might provide a clue. However, EXA is another new technology with its own collection of issues, and given that we're dealing with issues from immature code, I don't think it's wise to solve it by adding in more immature code without having a deep understanding of why it makes it work. Also, if it requires dropping patch 120 from xserver, we'll get that regression for XAA users.

Tim, if you can identify which config variable in /etc/gdm/gdm.conf to switch, then other users could test that. However, it's hard to change gdm behavior without causing a regression somewhere, so this likely contains some regression risk as well. But it might be worthwhile to identify as a workaround, if we can get several of us to validate it.